ALE-Agent

AI agent · Research agent · Research by Sakana AI

Autonomous optimization agent for hard computational discovery

ALE-Agent is a specialized optimization agent from Sakana AI that solves hard computational problems and discovers novel algorithms.

Agent specs

The full spec card (context, pricing per token, modalities and more) on AntSpace. Sign up free to see the full spec card.

Languages
English

Category: AI Agents

Competition Record
1st place in AtCoder Heuristic Contest 058
Key Technique
Inference-time scaling and self-learning
Primary Application
Hard optimization and algorithmic discovery
Research Lab
Recursive Self-Improvement (RSI) Lab
  • 1st place in AtCoder Heuristic Contest 058 (804 participants)
  • Autonomous discovery of novel algorithms
  • Specialized for NP-hard optimization problems
  • Features a self-learning mechanism from failures

Overview

ALE-Agent is Sakana AI's milestone optimization agent designed to tackle hard computational problems, such as NP-hard optimization tasks. It functions as an autonomous researcher that utilizes a self-learning mechanism to extract insights from trial-and-error failures, improving its problem-solving strategy over time. This technology serves as a core component of Sakana AI's research into recursive self-improvement and autonomous algorithmic discovery.

Optimization and Discovery

  • Inference-time scaling: The agent uses massive inference-time scaling to explore a wide range of potential solutions for complex problems.
  • Hard optimization focus: The agent is specifically designed to tackle hard optimization problems, such as NP-hard tasks, where true optima are computationally out of reach.

What it does

  • Generating novel algorithms
  • Solving hard optimization problems
  • Extracting insights from trial-and-error failures
  • Executing inference-time scaling for problem solving

Published results

  • 1st place out of 804 human participants in AtCoder Heuristic Contest 058
  • 21st out of 1,000 human participants in AtCoder Heuristic Competition (AHC) in May 2025

Use cases

  • Researchers employ the agent to tackle NP-hard optimization problems where true optima are computationally out of reach.
  • AI developers use the agent's self-learning mechanism to extract insights from trial-and-error failures for recursive self-improvement.

Specifications

Competition Record
1st place in AtCoder Heuristic Contest 058
Key Technique
Inference-time scaling and self-learning
Primary Application
Hard optimization and algorithmic discovery
Research Lab
Recursive Self-Improvement (RSI) Lab

FAQ

What is ALE-Agent?

ALE-Agent is a milestone optimization agent developed by Sakana AI that leverages massive inference-time scaling and self-learning to solve hard computational problems.

What technology powers ALE-Agent?

The agent uses a self-learning mechanism to extract insights from trial-and-error failures and employs massive inference-time scaling to find optimal solutions.

Added by AntSpace from public sources (sakana.ai).

Related

More AI agents by Sakana AI